Perplexity announced pplx-embed-v2-late on October 7, 2026: a pair of multimodal retrieval models, sized 0.6B and 9B. The company reports that the 9B model achieved 92.4% answer accuracy on its MADQA evaluation when paired with Gemini 3.5 Flash. The practical twist is that both models share an embedding space, allowing a 9B-built document index to be queried using the smaller 0.6B model.
What is pplx-embed-v2-late?
It is a family of multimodal, late-interaction retrieval models—not a general-purpose chatbot. Built on Qwen3.5 with bidirectional attention, the models encode text, images and visual documents into token-level representations for search. Perplexity describes the approach as ColBERT-style: each token produces a 128-dimensional vector, and query-document similarity is calculated using MaxSim. That preserves more token-level detail for retrieval than reducing an entire item to one pooled embedding. Perplexity’s release announcement describes the family, while the official Hugging Face model card documents implementation details.
As an Amazon Associate I earn from qualifying purchases.
The two checkpoints are pplx-embed-v2-late-0.6b and pplx-embed-v2-late-9b. Perplexity positions 0.6B for more latency-sensitive and edge deployments, and 9B as the larger, higher-quality option. The “0.6B” label does not mean all 594 million parameters are active for every encoding task: Perplexity says 240 million are active for text encoding and 340 million for image encoding.
Recommended Free Tools
Can the 0.6B model query an index built with 9B?
Yes. Perplexity says the two checkpoints share an embedding space, so documents can be encoded and indexed with 9B while live queries are encoded with 0.6B. This separates the heavier document-indexing work from the query path: the 9B encoding cost is paid when building the index, while query encoding uses the smaller model. Perplexity reports an average improvement of 1.6 percentage points across its domain-specific benchmarks for this asymmetric arrangement compared with using 0.6B on both sides. On ViDoRe v3 image retrieval, it reports 63.5% for 9B-index/0.6B-query versus 62.3% for 0.6B on both sides.
#1 Best Overall
![Arduino® UNO™ Q 4GB [ABX00173]- Hybrid Board, Qualcomm Dragonwing QRB2210 microprocessor (MPU) & STM32U585 Microcontroller(MCU), AI Vision, Voice, IoT, Robotics, Linux Debian OS, Wi-Fi 5, USB-C](https://m.media-amazon.com/images/I/51cl4xygwGL._SL160_.jpg)
- Dual-Brain Hybrid Power: Combines the Qualcomm Dragonwing QRB2210 MPU (Quad-core Arm Cortex-A53 @ 2.0 GHz CPU, Adreno GPU, AI acceleration) and the real-time, low-power STM32U585 MCU for advanced applications like object recognition, voice commands, and motion detection.
- AI & Linux Capabilities: Unlocks AI-powered vision and sound solutions; runs Linux Debian OS for coding in Python and supports the Arduino ecosystem with libraries and Sketches; quick start with Arduino App Lab.
- Advanced Features: Equipped with 4 GB LPDDR4 RAM, 32 GB eMMC built-in storage, ideal for single-board computer (SBC) mode, running multiple simultaneous high-level processes, more complex AI or ML models, extensive logs. Dual-band Wi-Fi 5 (2.4/5 GHz), Bluetooth 5.1, and high-speed headers for vision, audio, and display peripherals.
- Seamless Expansion & Connectivity: Features the classic UNO form factor for shields compatibility, an 8x13 LED matrix, and a Qwiic connector for easy expansion with Modulino nodes; power and connect via the USB-C connector.
- Intended Use & Development: The perfect platform for prototyping robotics or IoT projects, empowering innovators with a unified development experience to mix Arduino Sketches, Python scripts, and containerized AI models in a single interface.
Perplexity summarizes the capability this way: “Both models can be used independently, but the shared embedding space also allows the smaller model to query an index created with the larger model, combining higher-quality document embeddings with cheap, fast query-time inference.”
What does the 92.4% MADQA score mean?
It is Perplexity’s reported answer-accuracy result for the 9B retriever paired with Gemini 3.5 Flash—not a standalone score for the embedding model answering questions by itself. Perplexity says MADQA contains 500 human-authored questions over 800 heterogeneous real-world PDFs spanning more than 18,000 pages. The questions are designed to require evidence from the documents rather than general knowledge. The evaluation reports answer accuracy and page-level F1. Perplexity reports 90.1% MADQA answer accuracy for the 0.6B retriever in the described setup. These figures are company-reported; the cited materials do not establish an independent reproduction.
Rank #2
- Dual-Brain Hybrid Power: Combines the Qualcomm Dragonwing QRB2210 MPU (Quad-core Arm Cortex-A53 @ 2.0 GHz CPU, Adreno GPU, AI acceleration) and the real-time, low-power STM32U585 MCU for advanced applications like object recognition, voice commands, and motion detection.
- AI & Linux Capabilities: Unlocks AI-powered vision and sound solutions; runs Linux Debian OS for coding in Python and supports the Arduino ecosystem with libraries and Sketches; quick start with Arduino App Lab.
- Advanced Features: Equipped with 2 GB LPDDR4 RAM, 16 GB eMMC built-in storage, ideal to develop in PC-connected mode, running the OS, Python scripts, and basic network services (SSH) without a demanding GUI or heavy multitasking; great for lightweight AI and memory-optimized TinyML applications, needing local storage for basic OS and core libraries. Dual-band Wi-Fi 5 (2.4/5 GHz), Bluetooth 5.1, and high-speed headers for vision, audio, and display peripherals.
- Seamless Expansion & Connectivity: Features the classic UNO form factor for shields compatibility, an 8x13 LED matrix, and a Qwiic connector for easy expansion with Modulino nodes; power and connect via the USB-C connector.
- Intended Use & Development: The perfect platform for prototyping robotics or IoT projects, empowering innovators with a unified development experience to mix Arduino Sketches, Python scripts, and containerized AI models in a single interface.
Perplexity also reports ViDoRe v3 nDCG@10 results, which measure a different retrieval task and should not be treated as MADQA answer-accuracy scores:
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
| Checkpoint | ViDoRe v3 image nDCG@10 | ViDoRe v3 Markdown nDCG@10 |
|---|---|---|
| 0.6B | 62.3% | 61.2% |
| 9B | 65.2% | 64.7% |
These ViDoRe figures are reported by Perplexity in its model card; they are not directly comparable to MADQA’s answer-accuracy percentages.
Rank #3
- Single core ARM Cortex-A7 32-bit core, integrated with NEON and FPU
- Built in Micro's self-developed 4th generation NPU, with high computational accuracy and support for mixed quantization of int4, int8, and int16. Among them, int8 has a computing power of 0.5 TOPS and int4 has a computing power of up to 1.0 TOPS
- Built in self-developed 3rd generation ISP3.2, supports 4 million pixels, and supports various image enhancement and correction algorithms such as HDR, WDR, and multi-level denoisin
- It has powerful encoding performance, supports intelligent encoding, adapts to save bit rates according to the scene, and saves more than 50% of the bit rate compared to conventional CBR mode, making the captured images high-definition, smaller in size, and doubling the storage space
- The design with built-in RISC-V MCU supports low-power fast startup, 250ms fast capture, and simultaneous loading of AI model library, enabling facial recognition to be completed within 1 second
Which deployment approach fits?
The right arrangement depends on the quality target, query-time compute, indexing budget and whether data needs to stay local. Perplexity describes three main configurations:
| Documents indexed with | Queries encoded with | Best fit | Trade-off |
|---|---|---|---|
| 9B | 9B | Prioritizing the highest retrieval quality | Uses the larger model for both indexing and query encoding. |
| 0.6B | 0.6B | A more efficient, all-local setup | Uses the smaller model throughout; reported benchmark results are lower than 9B. |
| 9B | 0.6B | Improving document representations while keeping the smaller query path | Requires the added 9B document-encoding work during index construction. |
Perplexity also describes a local-cloud approach: use local 0.6B representations alongside results from a cloud-hosted 9B index, comparing or merging retrieval results. These are deployment options rather than guarantees of a particular latency, infrastructure bill or privacy outcome; those depend on the implementation and where the data and models run.
Rank #4
- 【POWERFUL ESP32‑S3 CONTROLLER】Built‑in Xtensa 32‑bit LX7 dual‑core processor, 512KB SRAM, 8MB PSRAM, 16MB Flash for stable AI voice computing and multitask processing.
- 【Preloaded Dual AI Platforms】Comespre-installed with complete Deepseek and OpenAI voice dialogue projects.Experience intelligent voice interaction instantly. (Note: OpenAI functionality requires your own API key.)
- 【STABLE WIRELESS & CLEAR AUDIO】Integrated 2.4GHz Wi‑Fi + Bluetooth 5 (LE); dedicated audio decoding module for natural, responsive voice interaction.
- 【USER‑FRIENDLY VISUAL & PLUG‑AND‑PLAY】2” TFT‑SPI color screen shows real‑time chat; modular design, no extra wiring, ready to use after setup.
- 【FULL LEARNING SUPPORT】45 programmable GPIOs, rich interfaces, online web tutorials, free technical support for beginners & developers.
What does the documented implementation require?
The model card documents use through Sentence Transformers’ MultiVectorEncoder, with separate query and document encoding calls and MaxSim similarity. It lists sentence-transformers >= 6.0.0 and transformers >= 5.4.0. According to the card, the exported model uses native Sentence Transformers modules and does not require custom Python code.
- The documented flow uses separate text-only and image-only batches; mixed text-plus-image inputs are not supported.
- The card warns that PyLate inserts query and document markers in a different position from the one this model expects.
- The 0.6B checkpoint is listed under the MIT license. Its model-card page does not list deployment by an inference provider.
Software dependencies, model files, licensing details and hosted inference availability can change; consult the current model card when implementing the model.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
